EngramLab and Harvey open source 100M+ token synthetic law firm

Harvey: Harvey is an AI platform purpose-built for legal and professional services, developing agents that handle end-to-end legal work including research, drafting, and matter execution with grounded citations. The company advances legal AI through open research on benchmarks, post-training, and agent capabilities. Here, Harvey partnered with EngramLab to open-source a synthetic law firm dataset, enabling evaluation of agents' ability to understand firm processes and past work product. EngramLab: EngramLab is an AI research company focused on memory, continual learning, and resource-efficient systems that dynamically learn from experience and compress knowledge for reuse. Co-founder Dan Biderman leads efforts in building models that address complex, context-dependent problems beyond standard retrieval methods. In this news, EngramLab collaborated with Harvey to create a synthetic law firm environment for testing AI agents on intricate legal tasks like tracking incomplete M&A deals across client files. Dan Biderman: Dan Biderman is CEO and co-founder of EngramLab, an AI researcher with a background in computational neuroscience and continual learning systems. He focuses on models that learn dynamically from experience to handle ambient, hard-to-search queries. In the news, he discusses the types of complex financing and deal-related problems EngramLab addresses in the context of the Harvey collaboration. Niko Gruppen: Niko Gruppen serves as Head of Applied Research at Harvey, overseeing teams that build legal agent benchmarks and synthetic datasets for training and evaluation. He brings experience from roles including Google Brain. In the news, he collaborated on the deep dive into the open-sourced synthetic law firm environment designed to test agents' comprehension of firm knowledge. Julio Pereyra: Julio Pereyra leads the applied legal research team at Harvey and contributes to AI agent development and benchmarking efforts. With prior experience as a lawyer, he helps bridge legal domain expertise with AI advancements. In this news, he co-authored a deep dive on the synthetic law firm project and its implications for agent evaluation. AI Collaboration: Partnerships between specialized AI labs like EngramLab and legal AI platforms like Harvey are driving innovations in agent capabilities for handling nuanced, multi-file legal queries that standard retrieval methods cannot address. Agent Evaluation: Recent efforts in the legal AI space emphasize building test environments that simulate a firm's historical work product to assess how well agents can inform current decisions based on past practice. Legal AI Research: Harvey continues to advance open research in legal AI by releasing benchmarks and synthetic environments that test agents on complex, knowledge-intensive tasks drawn from real firm practices.